Fixed spelling

[ROCm/rccl-tests commit: f74c04b686]
This commit is contained in:
corey-derochie-amd
2024-05-02 09:18:25 -06:00
کامیت شده توسط GitHub
والد 85bdda3812
کامیت fe151f517b
@@ -653,7 +653,7 @@ testResult_t TimeTest(struct threadArgs* args, ncclDataType_t type, const char*
if (cudaGraphLaunches >= 1) {
// Begin cuda graph capture
for (int i=0; i<args->nGpus; i++) {
// Thread local mdoe is needed for:
// Thread local mode is needed for:
// - Multi-thread mode: where graph capture and instantiation can happen concurrently across threads
// - P2P pre-connect: when there is no warm-up, P2P pre-connect is done during graph capture.
// Since pre-connect calls cudaMalloc, we cannot use global capture mode
@@ -703,7 +703,7 @@ testResult_t TimeTest(struct threadArgs* args, ncclDataType_t type, const char*
if (cudaGraphLaunches >= 1) {
// Begin cuda graph capture
for (int i=0; i<args->nGpus; i++) {
// Thread local mdoe is needed for:
// Thread local mode is needed for:
// - Multi-thread mode: where graph capture and instantiation can happen concurrently across threads
// - P2P pre-connect: when there is no warm-up, P2P pre-connect is done during graph capture.
// Since pre-connect calls cudaMalloc, we cannot use global capture mode